SlowMist Warns Malicious Code Hit MemoryOS and AI-Agent Packages
Affected releases include MemoryOS 2.0.34 on PyPI and three npm versions of an OpenClaw plugin, with possible prompt exposure.
SlowMist warned that malicious code was inserted into MemoryOS and a related AI-agent plugin, creating a software supply-chain risk for developers and crypto infrastructure teams using the packages.
The affected releases include MemoryOS 2.0.34, an open-source long-term memory library for large language models and AI agents distributed through PyPI. The compromised npm package, memtensor/memos-cloud-openclaw-plugin, connects to the OpenClaw runtime and was affected in versions 0.1.21, 0.1.23 and 0.1.25.
The malicious payload is a cross-platform Go binary that executes when an affected package is loaded or imported. The npm plugin may also expose users’ prompt contents.
SlowMist recommended removing the affected releases or downgrading to MemoryOS 2.0.33 and npm version 0.1.20. It also advised terminating the sckit process, blocking related infrastructure, reviewing network activity and rotating credentials in affected environments.
The incident highlights the risk of compromised developer tools being pulled into AI and crypto workflows through routine package installation or updates.
